RAPID: Understanding COVID-19 Transmission With Non-Markovian Models

With COVID-19, the world has experienced the most significant pandemic of contemporary history. In an effort to reduce the virus transmission, different mitigation strategies have been proposed and implemented. In this situation, models of disease transmission have emerged as a key tool to predict current and future characteristics of COVID-19 spreading, with or without the implementation of mitigation strategies, and to guide policymaking decisions. Models however are accurate predictors if they are built upon reliable data and evidence-supported assumptions. Large swings in model predictions can be imputed to assumptions not supported by data or evidence, with consequences on the model reliability. One typical assumption is the exponential distribution of the transition times of individuals between different states of disease (i.e., compartments that mark individuals as susceptible, exposed, infected, and recovered). However, recent observations of COVID-19 data, highlight non-exponential distributions for some critical transition times, such as the infectious period. This directly impacts the accuracy of the models. With this in mind, the goals of this project are to: 1) develop network-based compartmental meta-population models that accept arbitrary distributions for the transition times of the individual between different compartments; 2) develop rigorous methodologies to estimate unknown parameters of the model using stochastic optimization methods; 3) determine contact networks tailored for regions receiving lower attention, such as rural areas. Successful completion of this project will provide benefits to the USA public health, in particular to the analysis and monitoring of COVID-19. More accurate model-based testing of mitigation strategies will help public health officials to select strategies and to gather trust and support around mitigation policies. This way, health policymakers, modelers, and the general public will share common goals toward eventually stopping COVID-19.

In this project, the team will develop non-Markovian models that are driven by empirically determined distributions of transition times as suggested by recent results from analyzing data for COVID-19, highlighting the non-exponential distributions for some critical transition times. This novel aspect of our proposed model produces more accurate estimates of the current and future outbreak characteristics. For example, we can estimate the number of undetected infected people more accurately than models assuming the exponential distribution. Furthermore, as any model has known and unknown parameters, the estimation of the unknown parameters is critical to the model accuracy. To estimate the unknown parameters for this COVID-19 pandemic in the USA, we plan to use epidemic curves generated using the meta-population model and stochastic optimization techniques. The increased model accuracy produces better estimates of the outbreak characteristics and in turn better predictions of the mitigation policies effectiveness. Finally, any network-based model requires to input of a network representing contacts or movements. While these contact networks are available for some cities affected by the pandemic, rural regions have been less analyzed, despite occasionally being hot spots. The team will work to determine data-driven contact networks for certain rural areas of interest, for which we will apply and test our modeling approaches.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.